ECG based Arrhythmia Detection using Hybrid Deep Learning Model

Authors:

 Mrs. G.Bhagya Lakshmi, Majji Indu, Jonnadula Mohan Naga Vignesh, Jammu Rishitha.

Department of Information Engineering and Computational Technology, MVGR College of Engineering (A), Vizianagaram, Andhra Pradesh, India

Abstract — ECG remains one of the most reliable tools for early cardiac diagnosis arrhythmias in particular, where delayed detection can lead to serious clinical consequences. This work builds a hybrid deep learning model trained on the MIT-BIH Arrhythmia Dataset to automate that detection.

The architecture combines three components. CNN layers extract local morphological features from ECG waveforms. BR-SquareNet residual blocks refine those features while keeping training stable residual connections preserve original signal information and help the model catch subtle anomalies that deeper layers might otherwise lose. BiLSTM networks then handle the temporal side, modeling beat-to-beat variations across the signal and enabling accurate classification across multiple arrhythmia types. ECG data is preprocessed and structured for one-dimensional convolutional analysis before entering the pipeline, and fully connected dense layers with a Softmax classifier produce the final per-beat predictions.

The model is evaluated using accuracy, precision, recall, and F1-score, with confusion matrices and learning curves used to analyze performance in detail. Results show that combining spatial and temporal learning in a single architecture improves ECG classification over approaches that handle only one dimension. More practically, a system like this gives clinicians a faster and more consistent diagnostic tool, one that could meaningfully contribute to early identification of cardiac conditions in real clinical workflows.

Keywords: Arrhythmia Classification, Electrocardiogram (ECG), Deep Learning, Residual Networks, Time-Series Analysis.